In April 2012, an 18.9 m long core with frozen sediment (BK-8) was drilled from the top of an eroding Yedoma hill at 34 m a.s.l. (Fig. 1), located at about 100 m from the cliff edge on the western coast of the Buor Khaya Peninsula (71.420∘ N, 132.111∘ E). Detailed descriptions of the fieldwork and cryolithological properties (i.e. ground ice and sediment features) are published in Günther et al. (2013a) and Schirrmeister et al. (2016). Radiocarbon dating was performed on bulk plant macro remains by accelerator mass spectrometry at the CologneAMS laboratory, Germany (Schirrmeister et al., 2016) and radiocarbon ages are given in Table 1. Based on field observations, core descriptions and analytical datasets, Schirrmeister et al. (2016) subdivided the core into six cryolithological units (Table 1). Absolute ice-contents varied between 66 and 84 wt %, which is typical for Yedoma (Schirrmeister et al., 2011b, 2016). Plant remains were sparsely distributed throughout the core and were composed mainly of fine rootlets, grass fragments, small woody pieces and very few seeds and fruits.

In January 2014 the core segments were cut into two halves. One was stored as an archive and the second was subsampled. The opening of the core and the subsampling took place at the German Research Centre for Geosciences Helmholtz Centre Potsdam in the climate chamber at a temperature of -10 ∘C, where no genetic experiments were performed either before or after. The drilling mud was removed using a band saw. Approximately 3 mm of the surface was removed with a clean straight draw knife of 225 mm length (Wilh. Schmitt & Comp. GmbH & Co. KG, Germany). Before use, the draw knife was cleaned with 5 % Deconex-solution (Th. Geyer, Germany), rinsed in purified water, followed by DNA-ExitusPlus treatment (VWR, Germany) and rinsing in purified water. Finally, the draw knife was soaked in 96 % technical ethanol (Carl Roth GmbH & Co. KG, Germany) and flamed. Then, about 1 mm of the newly exposed sediment was removed with a small clean knife (Th. Geyer, Germany). Each knife was used for one draw only. Before use, the small knife was cleaned with 5 % Deconex-solution, rinsed in purified water, followed by DNA-Away^® (Carl Roth GmbH & ca. KG, Germany) treatment, and rinsed in purified water. Finally, each side of the knife was UV-irradiated at close distance for 10 min in a CL1000 ultraviolet crosslinker (UVP, USA), as recommended by Champlot et al. (2010).

The samples were drilled through the clean surface using a TCT hole saw with one tooth and an outer diameter of 25 mm (Esska.de GmbH, Germany). Before use, they were cleaned in the same manner as the small knives. The ends of the retrieved cylindrical sediment pieces were cut using sterile disposable scalpel tips. After cutting the first end, the scalpel tip was cleaned using Deconex, rinsed with purified water and Ethanol, and finally flamed to prevent contamination from one end to the other. The DNA samples were stored at -20 ∘C and the pollen samples at 4 ∘C. In total, 54 samples were drilled (approximately 3 to 4 samples per metre) for each kind of analysis. All 54 samples were analysed for sedaDNA while only 32 were processed for pollen analysis. The ice-wedge segment was not sampled for DNA analyses, because it was shattered into pieces, which were too small for an intact piece from the inside to be drilled out.

DNA isolation and PCR setup was performed in the palaeogenetic laboratory of the Alfred-Wegener-Institute Helmholtz Centre for Polar and Marine Research in Potsdam, Germany. This lab is dedicated to ancient DNA isolation and PCR setup and is located in a building devoid of any molecular genetics lab work. The lab is cleaned frequently by the researchers and subjected to nightly UV-irradiation. All laboratory work was performed in a UVC/T-M-AR DNA/RNA cleaner-box (BIOSAN, Latvia). DNA isolations and PCR setups were performed on different days using dedicated sets of pipettes and equipment. Further precautions to reduce contamination included UV-irradiation of 10 × buffer, BSA, MgSO4 and DEPC-treated water for 10 min in a UV crosslinker in thin-walled PCR reaction tubes approximately 1 cm below the UV light bulbs (similar to recommendations of Champlot et al., 2010).

Total DNA was isolated from approximately 5 g of frozen permafrost sediment using the PowerMax^® Soil DNA Isolation Kit (Mo Bio Laboratories, Inc. USA). For the initial lysis and homogenization step, 0.8 mg peqGOLD proteinase K (VWR, Germany), 0.5 mL 1M Dithiotreitol (VWR, Germany) and 1.2 mL C1 solution and samples were added to 15 mL PowerBead solution, vortexed for 10 min and incubated overnight at 56 ∘C on a nutating mixer (VWR, Germany) under gentle agitation. All following steps were carried out according to the kit manufacturer's recommendations, using 1.6 mL elution buffer and extending the incubation time to 10 min for the final elution. One extraction blank was included for each isolation batch of 11 samples and processed in the same way as the samples.

The PCR reactions were performed with the trnL g and h primers (Taberlet et al., 2007). Both primers were modified on the 5′ end by unique 8 bp tags which varied from each other in at least five base pairs to distinguish samples after sequencing (Binladen et al., 2007) and were additionally elongated by NNN tagging to improve cluster detection on the sequencing platform (De Barba et al., 2014). The PCR reactions contained 1.25 U Platinum^® Taq High Fidelity DNA Polymerase (Invitrogen, USA), 1 × HiFi buffer, 2 mM MgSO4, 0.25 mM mixed dNTPs, 0.8 mg Bovine Serum Albumin (VWR, Germany), 0.2 mM of each primer and 3 µL DNA in a final volume of 25 µL. PCRs were carried out in a TProfessional Basic thermocycler (Biometra, Germany) with initial denaturation at 94 ∘C for 5 min, followed by 50 cycles of 94 ∘C for 30 s, 50 ∘C for 30 s, 68 ∘C for 30 s and a final extension at 72 ∘C for 10 min. To trace possible contamination one no template control (NTC) was included in each PCR and treated identically to the samples and the extraction blanks. To check if the PCR was successful and whether the products matched the expected size, 2 % agarose (Carl Roth GmbH & Co. KG, Germany) gels were used .

For each sample, we pooled two positive amplifications for sequencing, under the condition that the associated NTCs and extraction blank were negative. The two pooled positive amplifications were purified using the MinElute PCR Purification Kit (Qiagen, Germany), following the manufacturer's recommendations. Elution was carried out twice with DEPC-treated ultra-purified water to a final volume of 40 µL. The DNA concentrations were estimated with the dsDNA BR Assay and the Qubit^® 2.0 fluorometer (Invitrogen, USA) using 1 µL of the purified amplifications. To avoid bias based on differences in DNA concentration between samples, they were pooled in equimolar concentrations. All extraction blanks and NTCs were included in the sequencing run, using a standardized volume of 10 µL, even though they were negative in the PCRs. The sequencing results of extraction blanks and PCR controls are reported in the Supplement S4. Library preparation and sequencing on the Illumina HiSeq platform (2×125 bp) were performed by the Fasteris SA sequencing service (Switzerland).

The sequence quality was checked using FastQC (Andrews, 2010) (Supplement S1, Fig. S3.1). Filtering, sorting and taxonomic assignments of the sequences were performed using OBITools (Boyer et al., 2016). Forward and reverse reads were aligned to produce single sequences using illuminapairedend. These sequences were assigned to their samples based on exact matches to their tag-combination using ngsfilter, followed by obigrep to exclude sequences shorter than 10 bp and obiuniq with which duplicated sequences were merged while keeping the information to which sample the sequences originally belonged. Rare sequences occurring with less than 10 read counts across the dataset were excluded as probable artefacts using obigrep. Sequence variants probably attributable to PCR or sequencing errors were excluded by obiclean (Boyer et al., 2016).

Two reference databases were used for taxonomic assignments as described in Epp et al. (2015): the first is based on the quality-checked and curated Arctic and Boreal vascular plant and bryophyte reference libraries (composed of 1664 vascular plant and 486 bryophyte species) published by Sønstebø et al. (2010), Willerslev et al. (2014) and Soininen et al. (2015); the second is based on the EMBL Nucleotide Database standard sequence release 117 (Kanz et al., 2005; http://www.ebi.acuk/embl/). The sequences were assigned to taxon names based on sequence similarity to each of the reference databases using ecotag. The nomenclature for the taxonomic assignment follows the NCBI taxonomy (Sayers et al., 2009). When the same taxonomic names were given more than once to different sequences we attached the affix MOTU (molecular operational taxonomic unit).

To further remove noise in the dataset, sequences occurring less than 10 times in a sample were excluded using R v. 3.0.3 (R Core Team, 2014). Only sequences that displayed a best identity value of 1.0 to an entry of a reference database were kept and assigned a taxonomic name (Supplement S5). Sequences which were assigned to cultivated plants or those highly unlikely to occur in the Arctic were excluded from our analyses as probable contamination (Supplement S6). These exotic DNA sequences were detected in almost all samples but contained mostly less than 1 % of the total number of sequence counts with best identity of 1.0 (Supplement S6). The highest contribution of exotic DNA sequence counts within a sample was at 2.85 m depth from Musaceae. As a conservative measure we excluded sequences assigned to the PACMAD-clade (including Muhlenbergia richardsonis), since they are identical with Zea mays and authentic sequences cannot be distinguished from probable contamination. Nevertheless, the excluded sequences comprised only a small proportion (3.3 %) of all Poaceae sequences and an under-representation would thus have only a minor impact. Sequences appearing in the extraction blanks and NTCs comprised 3.4 % of all sequence counts and were also excluded (Supplement S4). In most cases these sequences were only found in the extraction blanks (55.6 %) or NTCs (33.3 %) and not in the samples, while approximately 10 % were probably derived from wrong sample assignment through tag-jumps.

For pollen analysis, 32 samples were selected . From each sample approximately 3 g (wet weight) were taken for sample preparation. For fluid samples (unfrozen due to storage at 4 ∘C) 1 mL of sediment was taken using a syringe. Standard preparation following Faegri and Iversen (1989) including KOH, HCl and HF treatment was used to extract pollen, spores and algae from the sediment. For calculation of concentration of pollen a Lycopodium spore tablet (batch no. 1031; n=20 848 ± 1460) was added to each sample (Stockmarr, 1971). Pollen of terrestrial and aquatic plants as well as common spores and algae were analysed using a light microscope (Zeiss Axioskop 2) under 400–600 × magnification (Supplement S7). At least 300 pollen grains, spores and algae were identified in each sample following sample sizes in Andreev et al. (2011). Published pollen atlases (Beug, 2004; Kupriyanova and Alyoshina, 1972, 1978; Moore et al., 1991; Savelieva et al., 2013; Sokolovskaya, 1958) and a pollen reference collection at the Arctic and Antarctic Research Institute (Sankt-Petersburg) and the Alfred Wegener Institute were used for taxonomic identification of pollen and spores. Spores were determined according to van Geel (2001), van Geel et al. (1983), van Geel and Aptroot (2006). Freshwater algae were determined using Jankovská and Komárek (2000) and Komárek and Jankovská (2003).

Analyses for sedaDNA and pollen were carried out in the same manner. After first inspection of the data, we assigned Cyperaceae to the local component (Moore et al., 1991) of the swamp and aquatic taxa to get a better picture about changes in the terrestrial sedaDNA and pollen signals. We thus separated the data of both proxies into two datasets: (1) terrestrial and (2) swamp and aquatic. Recorded bryophytes for sedaDNA and spores were described, but since they were sparse, they could not be analysed statistically. Due to the higher taxonomic resolution of the sedaDNA, Kobresia (Cyperaceae) remained in the terrestrial sedaDNA set, while Poinae – either Arctophila fulva or Dupontia fisheri – were included in the swamp and aquatic set as both can occur within low-centred polygons (Aiken et al., 2007). Rarefaction curves were produced and rarefied taxon richness calculated using rarecurve and rarefy, to compare taxon richness at a particular number of sequences or pollen grains, determined by the lowest number of retrieved sequences or pollen grains among all samples (Heck et al., 1975; Hurlbert, 1971). This was performed for the terrestrial and swamp and aquatic datasets separately.

For statistical analyses, only taxa that were present in at least six samples in the terrestrial and in three samples in the taxonomically poorer swamp and aquatic datasets were included. Relative proportions of taxa within a sample were calculated on the basis of their sequence count or pollen sum in each dataset. A double square-root transformation was performed on the relative proportions of the sedaDNA dataset to mitigate the effect of over-represented and rare sequences, while square-root transformation was applied to the pollen dataset, since differences between counts were not as pronounced. A constrained hierarchical clustering approach (CONISS; Grimm, 1987) was performed with clusters constrained by depth using chclust. The relative proportions were thereafter transformed to Euclidean distances by vegdist. Zoning was guided by the broken-stick model (Bennett, 1996; MacArthur, 1957) using bstick, but with the condition that a minimum of five samples was necessary to assign a zone for the DNA datasets to avoid inflation of several very small zones. For pollen analysis a minimum of four samples was necessary to assign a zone, since fewer samples were taken compared to the DNA dataset. The stratigrams were plotted for each dataset separately with strat.plot.

Ratios of terrestrial to swamp and aquatic taxa and Poaceae to Cyperaceae were built (Mensing et al., 2008) to assess which contributed most to a sample, following Eqs. (1) and (2): ratio=terrestrial-aquaticterrestrial+aquatic,ratio=Poaceae-CyperaceaePoaceae+Cyperaceae. For this, the sums of sequence counts or pollen grains of (1) all terrestrial, (2) all swamp and aquatic taxa, (3) Poaceae and (4) Cyperaceae in a sample were built. The ratios range between -1.0 and 1.0 with negative values indicating dominance of swamp and aquatic or Cyperaceae sequences or pollen grains and positive values indicating dominance of terrestrial or Poaceae sequences or pollen grains, while a value of zero indicates an equal contribution of both.

A principal component analysis (PCA) was applied on the double square-root transformed relative proportions for sedaDNA and square-root transformed relative proportions for pollen using rda in order to portray the major structure in the multivariate dataset. Loadings for principle component 1 (PC1) and principal component 2 (PC2) axes were extracted and visualized in biplots. For better visibility, only terrestrial taxa that explained the most are plotted (n=20 for sedaDNA; n=25 for pollen).

The statistical analyses were performed in R v. 3.0.3 (R Core Team, 2014) using the packages “vegan” (Oksanen et al., 2011), “rioja” (Juggins, 2012) and “analogue” (Simpson, 2007; Simpson and Oksanen, 2016). The age–depth model was built using the Bacon package (Blaauw and Christen, 2011) in R. The paleogenetic and palynological datasets are deposited at 10.1594/PANGAEA.870897.

Ratios were built for the sedaDNA and pollen datasets to assess whether terrestrial or swamp and aquatic taxa contributed more sequences or pollen grains to a sample and were compared to the corresponding Poaceae–Cyperaceae ratios. This allowed us to trace local hydrological changes and to identify drier (positive values) and wetter phases (negative values) (Fig. 10). Generally, sedaDNA and pollen show similar trends for both ratios, with an exception between 11.7 and 12.12 m depth. The Poaceae–Cyperaceae ratio of the sedaDNA exhibits highly fluctuating ratios across the core and mostly follows the pattern of the terrestrial–aquatic ratio. A total of 18 sedaDNA samples are dominated by swamp and aquatic taxa: 4 above the ice wedge and 14 below. The pollen ratios show more moderate fluctuations and only two samples are dominated by swamp and aquatic taxa at 11.7 and 12.12 m depth. However, in six samples Cyperaceae dominate over Poaceae, one at 2.85 m and five between 11.7 m and 14.65 m, whilst samples between 13.0 and 14.0 m have equal contributions of Poaceae and Cyperaceae pollen.

All samples from the BK-8 sediment core contained plant-derived DNA and pollen. The two proxies are known to complement each other (e.g. Jørgensen et al., 2012), and differences in the obtained data result mostly from the spatial scale at which sedaDNA and pollen originate (local vs. regional signal) and technical biases, which lead to variations in the taxonomic richness, the level of taxonomic resolution and the strength of taphonomic processes in both proxies.

The different spatial scales of sedaDNA (local) and pollen (local to extra-regional) records are an important aspect of the differences identified in the taxon spectra and thus indicate complementarity rather than direct comparability of the proxies. In most of the samples we did not detect conifer-derived sedaDNA, although they are present in the pollen record. Hence, the Pinaceae pollen presumably originated from extra-regional stands (Birks, 2001; van der Knaap, 1987). Furthermore, the applied sedaDNA marker is located on the chloroplast genome, which is transmitted through pollen in Pinaceae (reviewed in Mogensen, 1996). If pollen contributed significantly to the sedaDNA record, we would expect to find it at least in samples with high Pinaceae proportions, which we did not. This supports two assumptions about sedaDNA: first, that sedaDNA originates mainly locally (Haile et al., 2007, 2009; Jørgensen et al., 2012; Parducci et al., 2013; Pedersen et al., 2016; Sjögren et al., 2016; Yoccoz et al., 2012) and, second, that it is predominantly derived from roots and other plant parts rather than from pollen (Jørgensen et al., 2012; Levy-Booth et al., 2007; Parducci et al., 2013; Pedersen et al., 2016; Sjögren et al., 2016; Willerslev et al., 2003). Overall, we find a steady dominance of Saliceae (which we interpret as Salix), Poaceae and Cyperaceae sequences across all samples of the core. This is likely caused by the huge below-ground biomass of these taxa in tundra environments, which can far exceed the above-ground biomass. According to Iversen et al. (2015), the ratio of below- to above-ground biomass in tundra is highest for sedges and grasses, followed by shrubs, and is lowest for forbs. Since Salix is also found throughout the pollen record, we assume that it was locally present throughout the investigated time frame. This further supports the general view that sedaDNA mainly presents a local signal.

Technical and taphonomic biases of pollen data are well known. For example, standard pollen sample preparation, as applied in this study, may (partly) destroy Luzula and Larix pollen grains (Moore et al., 1991). Hydrophytes are largely under-represented in the pollen dataset when compared to the sedaDNA results, which may be caused by low pollen production or insufficient sedimentation, as pollen from, e.g., Potamogeton tends to float on the water surface for pollination (Cox, 1988; Preston and Croft, 1997). While taphonomic biases in sedaDNA are still not well understood and part of ongoing research, especially for lake sediments (Alsos et al., 2015), the technical biases of sedaDNA are known and have been reviewed in Hansen et al. (2006), Schnell et al. (2015) and Thomsen and Willerslev (2015). We found an inflation of unique sequence types, attributable to PCR errors, across the whole dataset, except for those which we deemed as possible contamination. Otherwise, our dataset shows low probabilities for erroneous base calls indicated by high sequencing qualities. We therefore assume that the taxa included in the analyses are authentic. A more detailed technical evaluation can be found in Supplement S2.

Compared to the number of vascular plant taxa (58) and bryophytes (4) recorded by pollen analysis, the sedaDNA approach recorded a higher number of both vascular plants (134) and bryophytes (20). Next to technical biases and taphonomy, the lower number of taxa recorded by pollen can be explained by the sampling effect and the taxonomic resolution. The counts of sequences and pollen grains differed by several orders of magnitude, which is reflected in rarefaction curves of sedaDNA reaching saturation while those of pollen do not. This indicates that our sequencing depth was adequate for sedaDNA. The number of pollen counts was guided by pollen records for the Laptev region, usually ranging between 100 and 600 counts (Andreev et al., 2011). However, for future studies a higher sampling effort should be considered. The recorded richness also depends on the taxonomic resolution, which depends on the marker employed for sedaDNA. The resolution of the trnL P6 loop marker (Taberlet et al., 2007) allowed assignment of 78 % of the retrieved sequences to species or genus level, while 71 % of pollen, spores and algae were identified to a similar taxonomic level. This is in the range of other sedaDNA studies focusing on Arctic vegetation from permafrost sediments (Taberlet et al., 2007: 90 % up to genus level, Sønstebø et al., 2010: 83 % for the oldest and 68 % for the youngest sample, Jørgensen et al., 2012: 81 %, Willerslev et al., 2014: 80 %). SedaDNA analysis was able to resolve 21 sequence types from Poaceae and Cyperaceae on the species or genus level. Although some of the sequence types within these families cannot be resolved due to insufficient variation in the marker region, they cannot be distinguished by pollen analysis either (Birks, 2001). A higher resolution would provide a better estimate of the taxonomic richness and thus greater insight into local environmental conditions. Additionally, we found that bryophytes are highly under-represented in our datasets, despite being reported to be highly abundant, diverse and functionally very important members of modern polygonal landscapes (Zibulski et al., 2013, 2016). Epp et al. (2012) developed a marker for bryophyte metabarcoding with an approximately 10 % higher resolution than the trnL P6 loop. Their marker, however, had a low amplification success rate in late Pleistocene samples. They highlighted two probable causes, which might hold true for the trnL P6 marker. First, the main bryophyte biomass is typically found above-ground, whereas roots are suspected to contribute the majority of vascular plant DNA in soil (Levy-Booth et al., 2007; Willerslev et al., 2003; Yoccoz et al., 2012), and second, the presence of secondary metabolites may increase DNA degradation rates after cell lysis (e.g. Xie and Lou, 2009).

Lastly, sedaDNA richness depends on the completeness of the reference database (Jørgensen et al., 2012; Parducci et al., 2013; Pedersen et al., 2013). By using the Arctic–Boreal reference database (Soininen et al., 2015; Sønstebø et al., 2010; Willerslev et al., 2014), we were able to increase the resolution for many taxa, plus the EMBL database allowed us to examine our sequences for possible contamination by food or cultivated plants.

The major part of the core below the ice wedge encompasses sediments deposited before the Last Glacial Maximum (LGM). According to sedaDNA and pollen this part can be divided into two zones; however the boundary between these zones differs slightly between the terrestrial and swamp and aquatic datasets and lies between 15 and 17 m. The zonation matches the major structure presented in the PCA biplots of sedaDNA and pollen. The pollen record from the whole core portrays an open landscape at the regional scale. Below the ice wedge the core shows fluctuating ages associated with high standard deviations, which can be explained either by the radiocarbon dating method being at its limit or by reworking of the sediments.

The deepest part of the core from 18.9 to ∼ 16 m sedaDNA reflects a local terrestrial flora with low taxonomic richness, comprising taxa such as Plantago, Puccinellia and Potentilla, with Carex aquatilis as the only wetland plant retrieved at 18.15 m depth. The pollen record of this zone is characterized by high proportions of shrub, Poaceae and tree pollen, but with low proportions of Larix. As Larix pollen has a very limited dispersal capacity owing to its size, weight and low quantity (Niemeyer et al., 2015; Sjögren et al., 2008), Larix stands are inferred to have been in the regional vicinity of the coring site, possibly even closer than they are today. In contrast to sedaDNA, high proportions of Pediastrum, Botryococcus and Zygnema-type algae and highest proportions of Potamogeton pollen overall imply the presence of a shallow pond (Andreev et al., 2002; Kienast et al., 2005). Only in this zone do the sedaDNA and pollen records show such distinct differences. In the deepest 2.5 m the amount of redeposited pre-Quaternary pollen and spores is highest (up to ∼ 5 %) among all samples. According to the sedimentary and hydrogeochemical results presented in Schirrmeister et al. (2016), this part of the core implies an ancient active layer. An active layer is prone to disturbances such as erosion, cryoturbation (through seasonal thawing and refreezing) and potentially also grazing, all of which allow for redeposition of older material.

At depths from approximately 16 m until the ice wedge at 8.35 m, sedaDNA and pollen reveal high taxonomic richness. SedaDNA portrays high diversity among grasses and forbs including swamp and aquatic taxa and high proportions of Salix. In several samples the sedaDNA record is dominated by Cyperaceae and other swamp and aquatic taxa, with mostly negative values on the ratio plot, especially for the sedaDNA dataset. This indicates wet conditions on a local scale and probably enhanced organic matter accumulation, as cold and anoxic conditions reduce decomposition rates (Davidson et al., 2000). These findings are supported by the sedimentary, hydrogeochemical (Schirrmeister et al., 2016) and biomarker analyses (Stapel et al., 2016) performed on the same core. Schirrmeister et al. (2016) and Stapel et al. (2016) identified less decomposed organic matter at depths of 10 m and between 11.2 and 15 m from higher total organic carbon (TOC) content, higher hydrogen index, lower δ13C values and high concentrations of branched Glycerol dialkyl glycerol tetraether (br-GDGTs, microbial membrane compounds). The taxonomic composition of sedaDNA comprised typical taxa of low-centred polygonal depressions such as Stuckenia, Hippuris and Caltha palustris, indicating the presence of a shallow pond (Kienast et al., 2008). Intermittently, the absence of hydrophytes and increasing proportions of Poinae (Arctophila fulva/Dupontia fisheri) and Cyperaceae indicate times without a pond and hence point towards temporal fluctuations in the hydrology of the depression. The temporal scale on which these fluctuations occurred, however, cannot be assessed due to cryoturbation in the sediments and the large uncertainties of the dating results.

As shown by recent studies of low-centred polygons, microtopographical differences resulting from the moisture gradient between the dry uplifted ridge and the wet depression shape the local plant community (de Klerk et al., 2009, 2011; Zibulski et al., 2016). According to the sedaDNA results, Carex probably occupied the major part of the polygon, whereas the ridge was likely covered by Salix along with Poaceae and forbs (de Klerk et al., 2009, 2011, 2014; Minke et al., 2007, 2009; Teltewskoi et al., 2016). The transitional zone from the ridge to the depression is characterized by an increase in moisture and was probably occupied by taxa such as Carex, Eriophorum, Comarum and Pedicularis (de Klerk et al., 2009, 2014; Savelieva et al., 2013; Zibulski et al., 2016). However, vegetation surveys along transects through modern low-centred polygons and temporal reconstructions from short cores (de Klerk et al., 2009, 2011, 2014; Minke et al., 2007, 2009; Teltewskoi et al., 2016; Zibulski et al., 2016) show high proportions of Vaccinium, Ledum palustre, Empetrum nigrum and Betula nana ssp. exilis and therefore display a different composition of Holocene polygons in comparison to our findings. This indicates that Holocene polygon mires might differ from those of the pre-LGM.

The palynological record in this zone (16–8.35 m) comprises Potamogeton pollen, Pediastrum, Botryococcus and Zygnema-type algae, which supports the presence of a shallow pond (Andreev et al., 2002; Kienast et al., 2005). Artemisia, Dryas and Papaveraceae indicate overall dry environmental conditions with probably more severe winters than today, while Potamogeton indicates warmer summers (Kienast et al., 2001, 2005). Annual precipitation of less than 250 mm and rapidly falling temperatures in winter must have occurred to allow thermal cracking of the soil to keep the active layer sufficiently shallow for the formation of ice wedges and ridges that enclose low-centred polygons (Minke et al., 2007). Low relative pollen proportions of trees from extra-regional stands and shrubs with high proportions of grasses, sedges and forbs are consistent with other published pollen and macrofossil analyses in this time interval and region (Andreev et al., 2011; Kienast et al., 2001; Sher et al., 2005). The recorded pollen spectra from our core furthermore tally with studies from the central Laptev region, in which pollen records are dominated by Cyperaceae and Poaceae with a constant presence of Salix and high abundances of Artemisia and Caryophyllaceae for 55 to 40 kyr BP (Andreev et al., 2011, and references therein). The decreasing proportions of Larix pollen with decreasing depth presented here may point towards a retreat of Larix stands or a reduction in pollen productivity through unfavourable environmental conditions.

The permafrost core was drilled at the top of a Yedoma hill (Schirrmeister et al., 2016). Wind and rain probably eroded most of the Holocene deposits, resulting in a hiatus between the sample of the modern core top and the second sample at 0.25 m depth (11.1 cal kyr BP). The upper part of the core consists of sediments dated to the transition from the late glacial to the early Holocene (13.4–11.1 cal kyr BP). As emphasized in Andreev et al. (2011), records of the late glacial transition are rare because of active thermoerosion. Hence, our results provide valuable information about the vegetation history in this region and organic matter composition. The sedaDNA results imply profound changes after the LGM, which is displayed in the major structure of the terrestrial PCA biplot. First, the local taxonomic richness decreased strongly. Second, the taxonomic composition of the local flora changed towards shrub tundra and was mainly characterized by high proportions of Betula, Salix and Equisetum with a low diversity among Poaceae and forbs with only a subset of the formerly present Cyperaceae. Highly fluctuating proportions, especially between Poaceae and Cyperaceae, indicate fluctuating moisture conditions but not the presence of a pond. In relatively drier periods, indicated by positive values in the ratio plot, the organic matter comprises mostly Poaceae, which were represented by a different composition in comparison to the pre-LGM, with Agrostidinae and Poa MOTU2. During moister periods, indicated by negative values in the ratio plot, increased proportions of wetland plants such as Eriophorum, Equisetum and Ranunculus were recorded. The high proportions of Equisetum in this time interval are supported by our spore record as well as in the palaeogenetic study of Willerslev et al. (2014) and the palynological review of Andreev et al. (2011).

The pollen analysis shows that the same dominant taxa detected by sedaDNA characterize the area on a regional scale and implies shrub tundra with Salix, Betula and Alnus (Andreev et al., 2011). The equal relative proportions between pollen from trees and shrubs and pollen from forbs, grasses and sedges indicate climate amelioration during the early Holocene (Andreev et al., 2011) with increased humidity after the marine transgression (Kienast et al., 2001). Shrub pollen increased in the Laptev Sea region approximately at 9 kyr BP (Andreev et al., 2011), while in the Khorogor Valley near Tiksi an increase, especially of Betula pollen, of up to 60 % was already recorded at 11.54 ± 0.06 kyr BP (Grosse et al., 2007; Khg-11). In this study the increase is recorded at 11.4 ± 0.05 kyr BP and therefore matches well with the pollen data of the Khorogor Valley. The uppermost samples are dominated by Poaceae followed by Cyperaceae pollen and show high proportions of Alnus, Betula and Salix along with Ericales and Ranunculaceae but low proportions of Artemisia, reflecting the modern pollen spectrum (Andreev et al., 2011; CAVM Team, 2003). In contrast, the sedaDNA surface sample is characterized by Agrostidinae, Eriophorum MOTU1 and Carex aquatilis. This most likely reflects their root biomass in this sample. Taken together, both proxies reflect the tussock-sedge, dwarf-shrub tundra according to the division of the Circumpolar Arctic Vegetation Map (CAVM Team, 2003).